Skip to content

Latest commit

 

History

History
38 lines (28 loc) · 3.94 KB

features.md

File metadata and controls

38 lines (28 loc) · 3.94 KB

Features and host support

@cortexkit/opencode-anthropic-auth is the OpenCode 1/2 plugin and setup CLI, maintained as a fork of the original @ex-machina/opencode-anthropic-auth plugin. @cortexkit/pi-anthropic-auth is a native Pi extension: it replaces the implementation of Pi's built-in Anthropic provider while keeping its anthropic provider ID and model IDs. Both use @cortexkit/anthropic-auth-core for OAuth, account storage, quota, caching, relay and signing.

Support matrix

Capability OpenCode Pi
Claude Pro/Max OAuth Native pool shared by OpenCode 1 and 2 Separate native pool
Claustrum vault accounts Optional enrollment Optional enrollment
Provider integration OpenCode 1 fetch/request transform; OpenCode 2 registered provider source registerProvider("anthropic")
Default pool ~/.config/opencode/anthropic-auth-pool.json ~/.pi/agent/anthropic-auth-pool.json
Legacy migration source ~/.config/opencode/anthropic-auth.json ~/.pi/agent/anthropic-auth.json
Accounts, routing, cache and diagnostics controls /claude menu /claude menu
Quota refresh and killswitch edits Menu actions Summary only; no killswitch edits
Prime quota-window scheduler Status and on/off actions Status only
Lane start Menu action Not available
Quota sidebar OpenCode 1 TUI plugin Not available
Cloudflare relay HTTP; OpenCode 1 also supports WebSocket HTTP or WebSocket

Setup prepares the selected hosts and records which credential store they may use before sign-in. OpenCode 2 has its own login and transport boundaries; OpenCode 1's sidebar, host-local quota-feed publisher and deterministic client safety recovery are not OpenCode 2 features.

What CortexKit adds over the original plugin

CortexKit adds account and request controls beyond OpenCode's or Pi's single built-in login:

  • Fallback accounts and quota-aware routing: ordered OAuth accounts and API-key routes, minimum remaining-quota thresholds, and sticky-balanced session affinity. See accounts and the routing reference.
  • Cache controls: explicit, automatic or hybrid prompt-cache placement, optional keepalive, and opt-in quota-window priming. Cache controls explain the schedules and billed synthetic requests.
  • Quota visibility and limits: per-account status, OpenCode quota refresh and killswitch thresholds, the OpenCode 1 sidebar, and its optional sanitized quota header feed.
  • Model controls: adaptive reasoning summaries, effort levels, fast mode and OpenCode 1 safety fallback. See models and safety fallback.
  • Optional custody and transport: Claustrum is a local credential-vault service that can supply accounts instead of the local pool; a user-owned Cloudflare Worker can relay request bodies. See vault custody and transports.
  • Claude-compatible request shaping: identity and billing headers, final-body signing, tool replay and prompt-cache placement. See request rewriting.

What the integrations do

OAuth requests use the native pool or an enrolled vault. A host's fixed, non-secret activation marker only selects the plugin; it is never used as a credential. Local refresh, account state and request authorization belong to the native runtime. The plugin shapes outgoing requests for Claude OAuth and provides one /claude menu for account and request controls. Pi's provider replacement uses that same credential runtime without changing Pi's provider ID.

Implementation: OpenCode 1, OpenCode 2, Pi, and Pi menus.